This cross-sectional study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) guidelines. Data were obtained from the NHANES 2011–2014 cycles. NHANES employs a complex, multistage probability sampling design to select participants representative of the non-institutionalized US civilian population. Data collection conducted by trained personnel following standardized protocols. The survey protocol was approved by the National Center for Health Statistics (NCHS) Research Ethics Committee, and all participants provided written informed consent prior to data collection.

Study participants underwent comprehensive health interviews that collected sociodemographic information (including age, gender, race/ethnicity, education level, marital status, and annual family income) and health-related behavioral data (encompassing alcohol consumption, smoking status, and physical activity levels). The interview also gathered information about medical conditions such as hypertension, diabetes, and depression, along with anthropometric measurements including body mass index (BMI) and waist circumference. From the initial sample, we included participants at or over the age of 60 and excluded those with incomplete cognitive assessment data and those with missing dietary fiber intake information. The final analytical sample comprised 2,713 adults (Figure 1).

Total dietary fiber intake (g/day) was the primary exposure in this study, while vitamin E intake (mg/day) was investigated as a potential mediator. Dietary data were collected through two 24-h dietary recalls, with the first recall conducted in-person and the second recall completed via telephone 3–10 days later. All dietary interviews were conducted by trained dietary interviewers using the United States Department of Agriculture's Automated Multiple-Pass Method, which employs a five-step process to enhance the accuracy and completeness of dietary recall. Nutrient intake calculations were performed using the United States Department of Agriculture's Food and Nutrient Database for Dietary Studies. Daily nutrient intakes were calculated by averaging the values from these two 24-h dietary recalls.

Cognitive function was evaluated using four standardized neuropsychological tests and served as the outcome measure of this study. The Consortium to Establish a Registry for Alzheimer's Disease (CERAD) Word Learning Test was administered to assess verbal learning and memory (13). This test consisted of two components: the immediate recall test (CERAD.IRT) and the delayed recall test (CERAD.DRT). For the immediate recall, participants were asked to read aloud and recall a list of 10 unrelated words, with the sum of correct responses from three consecutive trials recorded. The delayed recall was conducted 8–10 min after the initial learning trials, with scores ranging from 0 to 10. Executive function was assessed using the Animal Fluency Test (AFT), which required participants to name as many animals as possible within 1 min, with one point awarded for each correct animal named (14, 15). Processing speed and executive function were evaluated using the Digit Symbol Substitution Test (DSST) (16), where participants were given 2 min to match symbols to numbers in 133 boxes according to a provided legend, with scores ranging from 0 to 133 based on correct matches.

Global cognitive function was quantified using a composite z-score, calculated by averaging the standardized scores of all four cognitive tests (DSST, AFT, CERAD.IRT, and CERAD.DRT). Each test score was standardized based on the sample mean and standard deviation (SD), with higher scores indicating better cognitive performance (15, 17).

Covariates were selected based on their potential confounding effects in the relationship between dietary fiber intake and cognitive performance. Age (years), gender (male, female), race (Mexican American, other Hispanic, non-Hispanic White, non-Hispanic Black, non-Hispanic Asian, and other racial groups), annual family income (<20,000, 20,000–75,000, and >75,000 dollars), education level (<9, 9–12, or >12 years), marital status (married/living with partner or living alone), and medical history (history of hypertension, diabetes, and depression) were examined as demographic and health-related covariates. Anthropometric and lifestyle factors including BMI (kg/m²), waist circumference (cm), daily energy intake (kcal), alcohol consumption, smoking status, physical activity, and dietary vitamin intake (vitamins B1, B2, B6, B12, C, D, and E) were also considered as potential confounders.

Physical activity was quantified using total metabolic equivalent (MET) minutes of moderate to vigorous work and leisure activities, calculated from NHANES MET scores by multiplying the frequency, duration, and intensity. A threshold of 600 MET minutes per week was used to distinguish between high and low activity levels, with high activity levels offering significant health benefits (18). Smoking status was categorized into three groups: never smokers (lifetime consumption of fewer than 100 cigarettes), current smokers, and former smokers (those who quit after smoking more than 100 cigarettes). Similarly, alcohol consumption was classified into three groups: never drinkers (lifetime consumption of fewer than 12 drinks), current drinkers, and former drinkers (those who quit after consuming at least 12 drinks in their lifetime) (19).

Statistical analyses were performed using EmpowerStats (https://www.empowerstats.net/en/, X&Y Solutions, Inc., Boston, MA, USA) and R software version 4.4.1. A two-sided P-value <0.05 was considered to be statistically significant. Continuous variables with normal distribution were expressed as mean ± standard deviation (SD), while those with skewed distribution were presented as median with interquartile range. Categorical variables were described as frequencies and percentages. Participants were categorized into quartiles based on dietary fiber intake (Q1, ≤ 10.9 g/day; Q2, 10.9–15.4 g/day; Q3, 15.4–21.2 g/day; Q4, >21.2 g/day). In between-group comparisons were conducted using one-way analysis of variance (ANOVA) for normally distributed continuous variables, Kruskal–Wallis H test for skewed continuous variables, and chi-squared tests for categorical variables.

Associations between dietary fiber intake and cognitive performance were examined using univariate and multivariate linear regression models. Results are presented as β coefficients with 95% confidence intervals (CI). Potential confounders were selected based on their associations with outcomes of interest or a change in effect estimates exceeding 10% (20). Multicollinearity was assessed using the variance inflation factor (VIF), with VIF ≤ 5 indicating acceptable collinearity levels (Supplementary Table S1) (21). Variables demonstrating multicollinearity with cognitive function scores and dietary fiber (VIF > 5) were excluded from analysis. The multivariate models were adjusted for gender, age, race, education level, annual family income, alcohol status, hypertension, diabetes, physical activity, depression, and intake of vitamins B1 and D.

To investigate potential non-linear relationships between dietary fiber intake and cognitive performance, we employed generalized additive models (GAM) with smooth curve fitting. Two-piece-wise linear regression models were used to examine threshold effects, and the optimal threshold was determined using a log-likelihood ratio test comparing one-line linear regression with two-piece-wise linear models. The 95% CI for the turning point was calculated using bootstrap resampling.

Mediation analysis was performed using the “mediation” package in R version 4.4.1 to assess whether vitamin E intake mediated the association between dietary fiber intake and cognitive performance. Models were adjusted for the previously mentioned covariates. A significant mediating effect was established when all of the following criteria were met: significant indirect effect, significant total effect, and positive proportion of mediator effect. Mediation effects were assessed using the non-parametric percentile bootstrap method with 1,000 resamples to generate 95% confidence intervals. Mediation was considered statistically significant when confidence intervals excluded zero. To test the robustness of analysis, we performed two-piece-wise linear regression and mediation analysis stratified by gender (male, female), hypertension (yes or no), diabetes (yes, no, or borderline), and depression (yes or no). Several steps were taken to address potential sources of bias: (1) selection bias: NHANES sampling weights were applied to maintain national representativeness; (2) measurement bias: standardized assessment protocols were followed by trained personnel; (3) recall bias: the Automated Multiple-Pass Method was used to improve dietary recall accuracy; and (4) confounding: comprehensive adjustment for sociodemographic, lifestyle, and health factors.

This cross-sectional study included 2,713 participants with a mean age of 69.4 ± 6.8 years (Table 1), categorized into four quartiles based on dietary fiber intake. Higher fiber intake was significantly associated with male gender, higher education level, better socioeconomic status, and healthier lifestyle behaviors including lower smoking rates and increased physical activity (all P < 0.001). Additionally, participants in the highest fiber intake quartile (Q4) demonstrated lower prevalence of hypertension (57.2 vs. 67.9%, P < 0.001) and diabetes (19.9 vs. 27.5%, P < 0.001) compared to the lowest quartile (Q1). The DSST scores showed a clear positive trend across quartiles, with Q4 participants achieving notably higher scores (49.0 ± 16.6) compared to Q1 (42.2 ± 17.4, P < 0.001). Similarly, the AFT results exhibited a gradual improvement from Q1 (16.8 ± 5.6) to Q4 (18.2 ± 5.7, P < 0.001). The composite Z-scores revealed a significant positive association with fiber intake, progressing from −0.2 ± 0.8 in Q1 to 0.1 ± 0.7 in Q4 (P < 0.001).

Multiple linear regression models were employed to investigate the association between dietary fiber intake quartiles and cognitive performance (Table 2). After comprehensive adjustment for potential confounders in the Adjust II model (including race, education level, annual family income, alcohol status, hypertension, diabetes, physical activity, depression, vitamin B1 intake, and vitamin D intake), the associations with CERAD.IRT and CERAD.DRT were attenuated and lost statistical significance (all P values >0.05). Most notably, both DSST and the composite z-score demonstrated consistent associations with dietary fiber intake across all quartiles compared to the reference group (Q1). For DSST, statistically significant improvements were observed across all higher quartiles: Q2 (β = 1.60, 95% CI: 0.22–2.97, P = 0.0230), Q3 (β = 1.84, 95% CI: 0.40–3.29, P = 0.0123), and Q4 (β = 2.76, 95% CI: 1.18–4.34, P = 0.0006). Similarly, the composite z-score showed consistent positive associations across all fiber intake quartiles: Q2 (β = 0.08, 95% CI: 0.01–0.15, P = 0.0221), Q3 (β = 0.09, 95% CI: 0.01–0.16, P = 0.0186), and Q4 (β = 0.12, 95% CI: 0.04–0.20, P = 0.0029). In contrast, AFT showed statistical significance only in the highest quartile (Q4: β = 0.76, 95% CI: 0.14–1.39, P = 0.0162), with Q2 and Q3 failing to reach significance (P = 0.1159 and P = 0.0956, respectively).

Two-piece-wise linear regression analysis revealed varying relationships between dietary fiber intake and cognitive function across different cognitive assessments (Table 3). The DSST scores demonstrated a significant non-linear association with an inflection point at 29.65 g/day of fiber intake (likelihood ratio test P < 0.001), while the composite z-score showed a threshold at 22.65 g/day (likelihood ratio test P = 0.018). Below threshold, dietary fiber intake exhibited a positive association with DSST (β: 0.18, 95% CI: 0.01–0.26, P < 0.0001), whereas above this threshold, the relationship became negative (β: −0.15, 95% CI: −0.29 to −0.02, P = 0.0265). Similarly, the composite z-score showed a positive correlation below its inflection point (β: 0.01, 95% CI: 0.00–0.01, P = 0.0004), but the association plateaued past this threshold (β: −0.00, 95% CI: −0.01–0.00, P = 0.9043). In contrast, the AFT (likelihood ratio test P = 0.078), CERAD.IRT (likelihood ratio test P = 0.213) and CERAD.DRT (likelihood ratio test P = 0.220) did not demonstrate significant non-linear relationships. Then we used a GAM with smooth curve fitting (Figure 2). All models were adjusted for gender, age, race, education level, household annual income, drinking status, hypertension, diabetes, physical activity, depression, and intake of vitamins B1 and D.

Subgroup analyses stratified by gender, hypertension status, depression status, and diabetes categorization (including borderline cases) revealed significant threshold effects between dietary fiber intake and both DSST scores (Supplementary Tables S2–S5) and composite z-scores (Supplementary Tables S6–S9). Notably, these non-linear associations were consistently observed across all examined subgroups, with no significant interaction effects detected (all P-interaction >0.05).

To further explore the underlying mechanisms linking dietary fiber intake with DSST scores and composite cognitive z-scores, we conducted mediation analysis. As shown in Table 4 and Figure 3, the indirect effect of dietary fiber on DSST score through vitamin E intake was 0.102 (95% CI: 0.062, 0.146; P < 0.0001), and for the composite cognitive z-score, the indirect effect was 0.098 (95% CI: 0.059, 0.143; P < 0.0001). The proportion of the total effect mediated by vitamin E was 86.82% for DSST (95% CI: 55.85%, 160.29%; P < 0.0001) and 85.04% for the z-score (95% CI: 54.94%, 140.62%; P < 0.0001). Subgroup analyses further indicated that this mediating effect remained significant across strata of gender (male, female), hypertension (yes or no), diabetes (yes or no), and among participants without depression. However, the mediation effect did not reach statistical significance in participants with depression or borderline diabetes, which is likely attributable to the limited sample size in these subgroups (depression: n = 225; borderline diabetes: n = 118).